Submitted in partial fulfillment of the requirements for the degree of Master of Science in Computer Science, School of Engineering and Technology

Research Studies Project Report (M.Eng.) - Asian Institute of Technology, 2009

Nowadays, robots can perform many tasks. One of their potential tasks is to bin picking and part placement. Now many industries could use robots to perform bin picking, to saves time and money. To perform the difficult task of picking the objects from a bin, first robot needs to recognize particular objects. For recognition of the objects a 3D model beneficial, for that purpose, people use many methods; some people use CAD software to create 3D models, but this is very costly and time consuming. some people are now using computer vision techniques, which are very cheap compared to human labor and more expensive sensors such as lasers. So in this research study I use stereo vision-based system to reconstruct a 3D model of objects. Stereo vision can help the robot to estimate depth of objects. If correct depth information is known then a 3D model of the object can be reconstructed correctly. There are two kinds of stereo vision methods, useful for reconstructing 3D model of the objects. Here I used dense stereo correspondence which is more accurate and detailed than sparse correspondence methods. I use the dynamic programming method in OpenCV to create disparity map and then extract one object and make depth map of that object. From the depth map I generated a 3D point cloud. After that, in the future work I will apply 2D Delaunay triangulation to create a 3D mesh and visualize the object using OpenGL or any other tool. If the reconstructed 3D model is correct, the robot will recognize the object and grasp it correctly.